@sperax/tool-giza

Verifiable AI agent analytics with ZKML proof visualization on-chain

Giza is an agent tool from SperaxOS, packaged headless so you can call it from any agent framework. It ships two things: the manifest — a JSON-Schema function definition a model can call — and the executor that runs the call against the real API.

There is no UI layer and no framework lock-in. It works anywhere TypeScript runs.

Install

npm install @sperax/tool-giza

Usage

Call it directly

import { gizaExecutor } from '@sperax/tool-giza';

const result = await gizaExecutor.invoke('getAgentOverview', {"chain":"<chain>","limit":1}, {
  messageId: 'msg-1',
});

console.log(result.content); // prose summary written for the model to read
console.log(result.state);   // typed data payload for your own UI

Give it to a model

import Anthropic from '@anthropic-ai/sdk';
import { GizaManifest, gizaExecutor } from '@sperax/tool-giza';

const client = new Anthropic();

const response = await client.messages.create({
  model: 'claude-opus-4-8',
  max_tokens: 1024,
  messages: [{ role: 'user', content: 'Ask something this tool can answer' }],
  tools: GizaManifest.api.map((api) => ({
    name: api.name,
    description: api.description,
    input_schema: api.parameters,
  })),
});

for (const block of response.content) {
  if (block.type !== 'tool_use') continue;
  const result = await gizaExecutor.invoke(block.name, block.input, { messageId: response.id });
  console.log(result.content);
}

GizaManifest.api is already in JSON-Schema form, so it maps onto any tool-calling API — Anthropic, OpenAI, the Vercel AI SDK, or an MCP server — without translation.

Every executor returns a BuiltinToolResult — { success, content, state }. content is prose written for the model to read; state is the typed data payload for your own code. Executors never throw: a failed call comes back as { success: false, content: '<reason>' }, so a network blip degrades the answer instead of crashing the agent loop.

Configuration

None. This tool calls a public API directly and needs no key or origin configuration.

Tool identifier

sperax-giza

API reference

getAgentOverview

Get an overview of on-chain AI agents deployed on Giza, including their status, total inferences, and proofs generated. Supports filtering by chain, status, and result limit.

| Parameter | Type | Required | Description | | --- | --- | --- | --- | | chain | string | no | Filter agents by blockchain (e.g. "ethereum", "starknet", "arbitrum") | | limit | number | no | Maximum number of agents to return (default: 10) | | status | active | inactive | pending | no | Filter by agent status: "active", "inactive", or "pending" |

getProofHistory

Get ZKML proof history showing verification status, proof type (Cairo/Noir/RISC Zero), duration, and on-chain transaction details. Great for visualizing proof verification trends.

| Parameter | Type | Required | Description | | --- | --- | --- | --- | | agentId | string | no | Filter proofs by specific agent ID | | limit | number | no | Maximum number of proofs to return (default: 20) | | proofType | cairo | noir | risc0 | no | Filter by proof system: "cairo", "noir", or "risc0" |

getModelPerformance

Get detailed performance metrics for a specific AI model including inference latency, proof generation time, accuracy, and historical inference volume chart data.

| Parameter | Type | Required | Description | | --- | --- | --- | --- | | modelId | string | yes | The model ID to fetch performance data for |

getProtocolAnalytics

Get protocol-wide analytics for the Giza ecosystem including total agents, proofs, inference counts, chain breakdown, and trend visualization data.

Takes no parameters.

Types

Shared types come from @sperax/agent-tools-core: BuiltinToolManifest, BuiltinToolResult, BuiltinToolContext, and the BaseExecutor class every tool executor extends.

Related

• @sperax/agent-tools-core — the tool contract
• All SperaxOS agent tools — tool-giza is one of many
• SperaxOS — the agent workspace these tools were built for

License

Apache-2.0